Load Balancing

A client wants one dependable service endpoint. Behind that endpoint, the service may run on many backends. A load balancer accepts traffic for the service and selects an eligible backend for each request or connection.

That small act of selection solves several problems at once. It lets you add capacity without teaching every client about every backend. It keeps backend addresses private or replaceable. It also gives the traffic path one place to apply health decisions.

The core loop is:

client → frontend → eligible backend pool → selected backend → response
                         ↑
                  health and capacity

The frontend is the address and protocol that clients use. The backend pool is the current set of service instances. Eligibility answers whether a backend may receive new work. The balancing policy chooses among the eligible backends.

Load balancing does not create capacity. It distributes traffic across capacity that already exists. It also does not make a broken application healthy. A bad health signal can send traffic to a broken backend or remove a good one.

What the load balancer can see

The information available to the load balancer depends on where it operates.

A layer 4 load balancer works with network and transport information such as IP addresses, ports, TCP, and UDP. It can distribute connections without understanding an HTTP path or header. This fits non-HTTP protocols and traffic that should pass through with limited application awareness.